Vector-valued image regularization with PDE's: a common framework for different applications

We address the problem of vector-valued image regularization with variational methods and PDEs. From the study of existing formalisms, we propose a unifying framework based on a very local interpretation of the regularization processes. The resulting equations are then specialized into new regulariz...

Full description

Saved in:
Bibliographic Details
Main Authors: Tschumperle, D., Deriche, R.
Format: Conference Proceeding
Language:English
Subjects:
Anisotropic magnetoresistance
Chemical elements
Color
Equations
Geometry
Image processing
Image restoration
Smoothing methods
Symmetric matrices
Tensile stress
Citations: Items that cite this one
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:We address the problem of vector-valued image regularization with variational methods and PDEs. From the study of existing formalisms, we propose a unifying framework based on a very local interpretation of the regularization processes. The resulting equations are then specialized into new regularization PDEs and corresponding numerical schemes that respect the local geometry of vector-valued images. They are finally applied on a wide variety of image processing problems, including color image restoration, in-painting, magnification and flow visualization.
ISSN:1063-6919
DOI:10.1109/CVPR.2003.1211415